Method for Efficiently Simulating the Information Processing in Cells and Tissues of the Nervous System with a Temporal Series Compressed Encoding Neural Network

1. A computer-implemented method for real-time bio-realistic simulation of a neural network, the method comprising: a) storing in memory definitions for a multiplicity of N distinct types of finite state machines (sector types, representing different sub-neural structure and processing), wherein each sector type has a number of data inputs, a number of data outputs, types of data inputs, types of data outputs, and a data compressor, wherein the N distinct sector types have different numbers of inputs, different numbers of outputs, or different manners of processing input signals to produce output signals; b) storing in memory a multiplicity of sectors, wherein each sector is an instantiation of one of the multiplicity of sector types, wherein each sector has an internal state represented by a compressed history of data input to the sector, wherein the compressed history is factorized into distinct historical time intervals; c) defining interconnections among data inputs and data outputs of the multiplicity of sectors to form a network of sectors; and d) emulating in parallel the network of sectors using parallel computers, wherein the emulating comprises, for each sector, i) processing data input to the sector to produce data output from the sector, wherein the processing comprises computing the compressed history of data input to the sector, storing the compressed history of data input to the sector in memory; and ii) communicating the data output to inputs of other sectors in accordance with the defined interconnections.

2. The method of claim 1 wherein the multiplicity of sectors contains more than 22500 sectors per neuron on average.

3. The method of claim 1 wherein computing the compressed input history comprises compressing the data input to the sector during a time interval.

4. The method of claim 1 wherein the data compressor Σ=σ T of a sector type T is defined to compress with constraint y(0)=T(Σ(0)), where y(0) represents current data output from the sector.

5. The method of claim 1 wherein the data compressor Σ=σ T of a sector type T is factorable into data compressors defined on non-empty subsets of data inputs to the sector.

6. The method of claim 5 wherein the internal state is factored into a Cartesian product of factors stored in memory only once for multiple instantiations of the sector, thereby reducing the amount of memory used to describe states of the sectors.

7. The method of claim 1 wherein emulating the network comprises initializing each sector in the multiplicity of sectors such that every sector of a given type is initialized in the same state.

8. The method of claim 1 wherein the data compressor of each sector type comprises a lookup table stored in memory, and wherein computing the compressed history of data input to the sector comprises accessing a lookup table for a sector type for the sector.

9. The method of claim 1 wherein the processing data input to the sector to produce data output from the sector processing comprises computing from the stored compressed history of data input to the sector the data output from the sector.

10. The method of claim 9 wherein computing the output from the sector comprises computing an output function Q subjected to constraints imposed by replication of a statistical property of an output function F.

11. The method of claim 1 wherein the internal state requires at most 48 bits of memory per synapse.

12. The method of claim 1 wherein N is equal to the product of a number of neuron types to be emulated and a minimum number of neuron compartments to be considered.

13. The method of claim 1 further comprising generating search-engine style dictionaries and inverted indices to provide access to the multiplicity of sectors stored in memory.

14. The method of claim 13 wherein generating search-engine style dictionaries and inverted indices comprises storing shortened sector state values and sector connectivity maps on in-memory lists, wherein the method further comprises accessing a data item in the in-memory lists.

15. The method of claim 14 wherein the accessing comprises accessing the data item by pointers.

16. The method of claim 14 wherein the accessing comprises accessing the data item sequentially.

17. The method of claim 1 wherein the sector performs processing using a discrete time interval of less than 100 ms.

18. The method of claim 1 further comprising combining multiple signals that travel simultaneously through a dendrite by adding the multiple signals arithmetically to obtain a sum, applying to the sum an evolution operator to obtain a resulting signal, and then replacing the resulting signal by one signal in a discrete and finite subset of a set of possible signals.